MINIMIZING DEFLECTION AND

BENDING MOMENT IN A BEAM

WITH END SUPPORTS

Samir V. Amiouny John J. Bartholdi, IIIJohn H. Vande Vate

School of Industrial and Systems EngineeringGeorgia Institute of TechnologyAtlanta, Georgia 30332, USA

December 3, 1991; revised April 3, 2003

Abstract

We give heuristics to sequence blocks on a beam, like books on a

bookshelf, to minimize simultaneously the maximum deflection and the

maximum bending moment of the beam. For a beam with simple supports

at the ends, one heuristic places the blocks so that the maximum deflection

is no more than 16/9√

3 ≈ 1.027 times the theoretical minimum and the

maximum bending moment is within 4 times the minimum. Another

heuristic allows maximum deflection up to 2.054 times the theoretical

minimum but restricts the maximum bending moment to within 2 times

the minimum. Similar results hold for beams with fixed supports at the

ends.

Key words: combinatorial mechanics, heuristics, sequencing, beam, de-

flection, bending moment

1

1 Introduction

The limiting factor in the design of beams loaded by weights is often either

the permissible deflection [6] or the bending moment [8]. For given loads, the

positions at which they are placed determine the deflection and bending moment

along the beam; therefore deflection and bending moment can be controlled to

some extent by judicious sequencing of the loads. Unfortunately, as we shall

show, it can be computationally difficult to determine the best sequence of loads

on the beam; however we give fast heuristics that position the loads so that both

the maximum deflection and the maximum bending moment are guaranteed to

be not “too much” larger than the minimum possible. The guarantees are

analytical, not experimental, and so might be useful in certifying performance

of beams.

We model the problem of minimizing deflection of a beam as that of se-

quencing n homogeneous “blocks” (intervals) on a beam of length L so that the

maximum deflection at any point along the beam is as small as possible. We

make the simplifying assumption that any interaction between the blocks is neg-

ligible (as would be the case if the beam deflected only slightly or if the blocks

were not very high). We use the same model for the problem of minimizing

bending moment, except that the objective is to make the maximum bending

moment as small as possible. The jth block is characterized by its length lj and

weight wj . We assume that the blocks fill the beam exactly, possibly by the

artifice of including some imaginary blocks of zero weight.

The beam is initially straight, with a uniform cross section of area moment

of inertia I. We also make the usual assumptions of engineering design that

the beam material is isotropic and homogeneous, and that it obeys Hooke’s law

with a modulus of elasticity E [8].

We will discuss the case in which the beam has simple supports at both ends,

but our analysis also applies, with differences only in detail, when the beam has

fixed supports.

The objective of minimizing the maximum deflection is not identical to that

2

of minimizing the maximum bending moment. For example, consider a beam of

length 1 on which are to be placed two blocks of weight w1 = w2 = 1 and length

l1 = l2 = 0.1 and a third block of weight w3 = 1.02 and length l3 = 0.2. Then

to minimize the maximum deflection one must place blocks 1 and 3 at one end

of the beam and block 2 at the opposite end; but to minimize the maximum

bending moment one must place blocks 1 and 2 at one end of the beam and

block 3 at the opposite end. Thus the sequence that minimizes one objective

can fail to minimize the other. Nevertheless, these two objectives appear to be

highly coincident. Evidence of this is that each of our heuristics is guaranteed

to perform “well” with respect to both objectives simultaneously—even though

the point at which maximimum deflection is achieved can be distance nearly L/2

from the point at which maximum bending moment is achieved. Furthermore,

for every result we prove about deflection, there is a similar result about bending

moment that is provable by a similar argument. (Accordingly we give detailed

arguments only for deflection.)

All of our heuristics work by reducing the deflection or the bending moment

at the center of the beam. Fortunately, as we show, neither the maximum

deflection nor maximum bending moment can be much greater than that at the

center, regardless of the placement of the blocks.

2 Deflection and bending moment

Standard engineering design textbooks catalogue equations for the deflection

and bending moment of a beam with different types of supports [7, 8]. For

example, consider a beam of length L on which is superimposed a coordinate

axis with the origin at the leftmost end as in Figure 1. If the beam has simple

supports at the ends, then the deflection at any point x due to a point force of

magnitude F applied at xF is

D(x, F, xF ) =

F (L−xF )x(2LxF−x2−x2

F )6EIL , for x < xF ;

D(L− x, F, L− xF ), for x ≥ xF ,(1)

3

40

4L

•L/2

.............................................................................................................

..................

........

..................

F

xF.............................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................

...............................................................................................................................................................

......................................................................................

Figure 1: Application of a point force to a beam.

where the second term follows by symmetry of the beam and supports. The

bending moment is

M(x, F, xF ) =

F (L−xF )x

L for x < xF ;

M(L− x, F, L− xF ) for x ≥ xF .(2)

(Strictly speaking we have written the negative of the deflection. We take this

liberty for convenience of presentation so that we can speak of “minimizing

the maximum” for both deflection or bending moment. The alternative is to

speak of “maximizing the minimum” deflection and “minimizing the maximum”

bending moment, with obvious opportunities for confusion.)

We can compute the deflection and bending moment due to a block of length

l ≤ L and with homogeneous weight distribution of w/l units of weight per unit

length by invoking the following.

The Principle of Superposition The deflection (bending moment) at any

point in a beam subject to multiple loading is equal to the sum of the

deflections (bending moments) caused by each load acting separately [8].

We use Equation 1 to calculate the deflection due to an infinitesimal section

of length dl, and then, by the Principle of Superposition, we integrate that

expression over the length of the block (Figure 2). Then, when the block is

placed with its center at r, the deflection at any point x is

D(x,w, l, r) =

(−L+r)wx(4r2+l2−8Lr+4x2)24EIL , for x ≤ r − l/2;

D(L− x,w, l, L− r), for x ≥ r + l/2;

D(x, w(x−r+l/2)l , x− r + l/2, x+r−l/2

2 )

+ D(x, w(r+l/2−x)l , r + l/2− x, r+l/2+x

2 ), otherwise

4

40

4L

•L/2

•r

.

.

.

.

.

.

.

.

.

.

.

.

.

.

dl

......................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................

..............................................................................................................................................

..............

Figure 2: Deflection at the center of a beam of length L due to a homogeneousblock.

where the second expression follows by symmetry of the beam and supports, and

the last expression follows from the Principle of Superposition, which allows us

to write the deflection due to a point underneath the block as the sum of the

deflections due to the portions of the block to the right and to the left of the

point. Note that we have written the deflection in such a way as to emphasize

that it is a function of the weight, length, and placement of the block. Also,

rather than simply writing its algebraic form, we have written the function

recursively to show its structure. Finally, we use D to refer to deflection due to

either point forces or blocks and rely on context to make the distinction clear.

A similar argument shows that the bending moment due to a homogeneous

block is

M(x,w, l, r) =

(L−r)wxL , for x ≤ r − l/2;

M(L− x,w, l, L− r), for x ≥ r + l/2;

M(x, w(x−r+l/2)l , x− r + l/2, x+r−l/2

2 )

+ M(x, w(r+l/2−x)l , r + l/2− x, r+l/2+x

2 ) otherwise.

It is straightforward to show that, for both point forces and blocks, deflection

is a concave function of x. Therefore, by the Principle of Superposition, and

because sums of concave functions are concave, the deflection due to a set of

point forces or blocks must be a concave function of x. Similarly, bending

moment is a concave function. Accordingly, for a set of point forces or blocks

whose positions are fixed, the point of greatest deflection or of greatest bending

5

moment can be found by an efficient one-dimensional search procedure such as

Fibonacci search [2].

3 The center of the beam

When a force is applied to a beam, the maximum deflection in the beam gener-

ally occurs elsewhere than at the point of application. Intuitively, one expects

the deflection at the center of the beam to be large, though not maximal. In

fact, for any placement of the blocks, neither the maximum deflection nor the

maximum bending moment can exceed that at the center by much:

Theorem 1. For a simply supported beam, the deflection at any point is at

most 16/9√

3 ≈ 1.027 times the deflection at the center of the beam and the

bending moment is at most 2 times that at the center.

Proof. We first prove the theorem for point forces and then argue that it must

hold for continuous loads as well.

Assume without loss of generality that a point force F acts to the right of

the center of the beam, so that xF ≥ L/2. Differentiating the first part of

Equation 1 with respect to x and setting the derivative to zero, we get the point

of maximum deflection in the simply supported beam. Substituting back in the

original equation, we get the value of the maximum deflection:

maxx

D(x, F, xF ) =FL2xF

9EI

(1− xF

L

) (2− xF

L

) √xF

3L

(2− xF

L

).

Taking the derivative of maxx D(x, F, xF )/D(L/2, F, xF ) with respect to xF ,

we find that this ratio is minimal at xF = L/2, where it assumes the value 1.

Furthermore, maxx D(x, F, xF )/D(L/2, F, xF ) increases with xF , so that the

ratio approaches its maximum value as the point of application of F approaches

the end of the beam. Evaluating this limit gives a value of 16/9√

3 ≈ 1.027.

Consider now a collection of point forces F1, . . . , Fn acting at (possibly)

different points on the beam, and let x∗ be the point of maximum deflection in

the beam. By the Principle of Superposition, the total deflection at the center is

6

∑ni=1 D(L/2, Fi, xi), and the maximum deflection is

∑ni=1 D(x∗, Fi, xi). Since

maxx D(x, Fi, xi) is the maximum deflection in the beam due to Fi alone, then

obviously D(x∗, Fi, xi) ≤ maxx D(x, Fi, xi). Summing the last inequality over

all the forces, we get

n∑i=1

D(x∗, Fi, xi) ≤n∑

i=1

maxx

D(x, Fi, xi) ≤16

9√

3

n∑i=1

D(L/2, Fi, xi).

Finally, when the beam is loaded with blocks rather than point forces, the

same argument holds with summation replaced by integration.

To establish the result for bending moment, we use a similar argument, but

base it on the fact that for a single force, the maximum bending moment occurs

at the point of application of the force, and has a magnitude of

maxx

M(x, F, xF ) = FxF

(1− xF

L

).

In this case the ratio maxx M(x, F, xF )/M(L/2, F, xF ) approaches its maximum

value of 2 as xF approaches L.

4 V-shaped sequences

A simple class of sequences reduces both deflection and bending moment at the

center of the beam and therefore tends to reduce the maximum values along

the entire beam. We call this class the V-shaped sequences: Each such sequence

has the property that all the blocks whose centers fall on the same side of the

center of the beam are arranged in non-decreasing order of average density wi/li

from the center towards the ends of the beam (and if the center of a block is

coincident with the center of the beam, then that block must be less dense than

one of its adjacent neighbors). We will prove that such sequences do not cause

“too much” deflection or bending moment at the center; but first we need the

following technical result.

Lemma 1. The farther a block is from the center of the beam, the smaller is

the deflection and the bending moment at the center due to that block.

7

Proof. When the block is completely to the left of the center of the beam, the

derivative of D(L/2, w, l, r) with respect to the distance r is, by simple algebra,

positive for 0 ≤ r ≤ L/2. Thus decreasing r, or by symmetry increasing r, away

from L/2 will decrease the deflection at the center.

If the block overlaps the center of the beam, we can assume without loss of

generality that the center of the block is to the right of that of the beam. Then

moving the block to the right a distance δ away from the center is equivalent

by the Principle of Superposition to cutting an imaginary section of length δ

from the left side of the block and placing at its right end. By the previous

discussion, this will reduce the deflection at the center.

A similar argument establishes the result for bending moment.

The following result says that the class of V-shaped sequences includes any

sequence that minimizes deflection or bending moment at the center of the

beam. This will be useful when we bound the quality of a V-shaped sequence.

Lemma 2. Any sequence that minimizes deflection at the center or that mini-

mizes bending moment at the center must be V-shaped.

Proof. By a simple interchange argument any sequence that is not V-shaped can

be improved: Within any sequence of blocks that is not V-shaped, there must

be an adjacent pair of blocks Bi and Bj that are in strictly decreasing order

of density; that is, wi/li > wj/lj and either ri < rj < L/2 or L/2 < ri < rj

(without loss of generality we assume the second case). By the Principle of

Superposition, the deflection at the center due to blocks Bi and Bj is equal to

that of two imaginary blocks Bu and Bv, shown in Figure 3, with lu = li + lj ,

lv = li, wu/lu = wj/lj , wv/lv = wi/li − wj/lj , ru = (ri + rj)/2, and rv = ri.

Interchanging Bi and Bj changes the deflection of the beam in exactly the same

way as would keeping Bu fixed and moving Bv outward by a distance lj . By

Lemma 1, this reduces the deflection at the center.

A similar argument establishes the claim for bending moment.

The following shows that no V-shaped sequence can cause “too much” de-

8

•L/2

Bi Bj

•L/2

Bu

Bv

Figure 3: The deflection in the beam due to blocks i and j, with heights pro-portional to their densities, is the same as that caused by blocks u and v.

flection or bending moment at the center of the beam. This will form the basis

of our heuristics.

Theorem 2. For any V-shaped sequence of a given set of blocks, the deflection

at the center of a beam is never more than twice the minimum possible and the

bending moment is never more than twice the minimum.

Proof. First we show that it is sufficient to consider only those cases in which all

blocks are of equal length. To see this, consider a set of n blocks for which the

worst V-shaped sequence produces a deflection DV1 at the center of a given beam,

and for which the optimal sequence produces a deflection D∗1 at the center of that

beam. Now imagine cutting those blocks into a set of equal length pieces, using,

for example, gcd(l1, l2, . . . , ln) as the common length (where gcd is the greatest

common divisor function). Let DV2 and D∗

2 be the deflections at the center of the

beam produced by the worst V-shaped sequence and by an optimal sequence

of the new set of blocks, respectively. Since the sequence that produced DV1

remains V-shaped when the blocks are cut, DV1 ≤ DV

2 ; and since the imaginary

blocks can be arranged more freely, D∗1 ≥ D∗

2 . This implies DV1 /D∗

1 ≤ DV2 /D∗

2 ,

and the worst V-shaped sequence for the imaginary set of equal length blocks

has no better performance than the worst V-shaped sequence for the original

set of blocks.

Now consider a set of n blocks of equal length L/n. For convenience, assume

the blocks are indexed so that w1 ≥ · · · ≥ wn. Then the V-shaped sequence

9

4 4.........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................

..............................................................................

..............................................................................

..............................................................................

..............................................................................

............................................................................................................

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Figure 4: The distribution of mass density in the worst V-shaped sequence.

that produces the greatest deflection at the center of the beam is

SV =

(B1, B2, . . . , Bn/2, Bn, Bn−1, . . . , Bn/2+1) for n even;

(B1, B2, . . . , B(n+1)/2, Bn, Bn−1, . . . , B(n+3)/2) for n odd.

These sequences are V-shaped, as suggested by Figure 4. That they are the

worst V-shaped sequences follows by an interchange argument: In any other V-

shaped sequence S of the blocks, B1 has to be at one of the ends of the sequence

S. Compare S with SV by comparing the positions of the successive blocks B2,

B3, . . . ; let Bj be the first of these blocks out of sequence in S and let Bk be the

block in S in the position of Bj in SV . Then j < k so block Bk is lighter than

block Bj and block Bj must be the block at the opposite end of S from B1. In

fact the blocks Bj+1, Bj+2, . . . , Bk−1 that are lighter than Bj but heavier than

Bk must also be on the same side as Bj so that S has the following structure.

S = (B1, B2, . . . , Bj−1, Bk, Bk−1, Bk−2, . . . , Bj+1, Bj).

Now consider two cases.

Case 1. j > k − j In this case, block Bk is closer to the center of the sequence

S than any of the blocks Bk−1, Bk−2, . . . , Bj . If we remove from each

block Bk−i, i = 1, 2, . . . , k− j weight Bk−i−Bk−i+1 and place it on block

Bk, we create the V -shaped sequence:

S′ = (B1, B2, . . . , Bj−1, Bj , Bk, Bk−1, . . . , Bj+1),

and since all the weight has been moved closer to the center, S′ produces

a larger deflection at the center than S according to Lemma 1.

Case 2. j ≤ k − j In this case, block Bk is farther from the center of the se-

quence S than some of the blocks Bk−1, Bk−2, . . . , Bj . We exchange

10

4 4................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................

..........................................

..........................................

..........................................

..........................................

..........................................

..........................................

..........................................

..........................................

..........................................

............................................................................................................................................................

Figure 5: The distribution of mass density after the blocks have been split intwo and arranged in an optimal sequence.

blocks B1, B2, . . . , Bj−1 with blocks Bj , Bj+1, . . . , B2j−2 to create the

V -shaped sequence:

S′ = (Bj , Bj+1, . . . , B2j−2, Bk, Bk−1, Bk−2, . . . , B2j−1, Bj−1, Bj−2, . . . , B1)

that produces the same deflection at the center as S. Now, Bj is the first

block that is out of sequence in S′ and block B2j−1 is the block that is in

its place. So, we may apply the arguments of Case 1 to the sequence S′

to create a V -shaped sequence S∗ that produces a larger deflection at the

center than S′ and hence also larger than S.

Now the problem of bounding the performance of a V-shaped sequence is

reduced to that of bounding the ratio of the deflection at the center of the beam

due to sequence SV to that due to sequence S∗. In sequence SV , block j weighs

wj and is distance rj = (2j−1)L/(2n) for j ≤ n/2 and rj = (n+2j−1)L/(2n) for

j > n/2 from the center of the beam. Substituting these values in the expression

for deflection gives a linear form in the wj , which we write as∑n

j=1 bjwj .

Rather than evaluate the deflection for the optimum sequence S∗, it is con-

venient to use the lower bound on deflection that we get by splitting each block

into two equal parts, indexed as j and n + j, and placing the pieces sym-

metrically about the center in a V-shape as suggested by Figure 5. Blocks

j and n + j each weigh wj/2 and are centered at rj = (2j − 1)L/(4n) and

rn+j = (4n − 2j + 1)L/(2n). Substituting these values in the expression for

deflection gives a linear form in the wj , which we write as∑n

j=1 cjwj . The

ratio R of the deflection at the center of the beam due to sequence SV to that

due to sequence S∗ has the form

R =

∑nj=1 bjwj∑nj=1 cjwj

≤ maxj

bj

cj,

11

and by tedious but simple algebra

maxj

bj

cj≤ 12n2 − 16j2 + 16j − 8

6n2 − 2j2 + 2j − 1

Taking the derivative of the last expression with respect to j, we see that it is

decreasing in j for j ≥ 1, and therefore is largest at j = 1 where

R ≤ 12n2 − 86n2 − 1

≤ 2 for n ≥ 1.

A similar argument establishes the bound for bending moment.

The bounds of 2 on the performance of arbitrary V-shaped sequences are

tight as can be seen from the following example: Assume L = 1 and consider

three blocks, of lengths l1 = l2 = l < 1/2, l3 = 1− 2l and of weights w1 = w2 =

1, w3 = 0. Then the ratios of deflection and bending moment at the center due

to the V-shaped sequence (B1, B2, B3) to those of the sequence (B1, B3, B2)

approach 2 as l approaches 0.

Invoking Theorem 1, we have the following guarantee of the quality of any

V-shaped sequence.

Corollary 1. Any V-shaped sequence produces a maximum deflection no more

than 32/9√

3 ≈ 2.054 times the theoretical minimum and a maximum bending

moment no more than 4 times the theoretical minimum.

Any algorithm that generates V-shaped sequences will inherit the corre-

sponding performance guarantees. There are several natural, simple algorithms

to generate V-shaped sequences. Among the more interesting is to sort the

blocks by density and then iteratively place the next densest block as far as

possible from the center of the beam. This requires O(n log n) effort due to the

sorting. While we have proved only the bounds 2.054 and 4, we suspect this

heuristic in fact has a stronger performance guarantee. We conjecture that any

sequence of blocks constructed by this heuristic produces deflection and bending

moment at the center that is no more than 5/4 times minimum. This would

mean that the maximum deflection would be no more than 20/9√

3 ≈ 1.283 and

the maximum bending moment would be no more than 5/2 times optimum.

12

5 Minimizing deflection at the center

A V-shaped sequence has the advantage of being easy to compute, but it does

not have a strong guarantee of quality because it only reduces deflection and

bending moment at the center of the beam; it does not minimize either of

them. With more effort we can compute sequences that exactly minimize the

deflection or the bending moment at the center of the beam. Such an algorithm

will capitalize more effectively on the bound of Theorem 1.

Our solution is via a dynamic programming recursion based on Lemma 2.

For convenience we assume that the lengths of the beam and of the blocks are

integral. We begin by sorting the blocks and relabelling them in non-decreasing

order of average density, so that w1/l1 ≤ · · · ≤ wn/ln. Now let D∗j (x) denote

the minimum deflection at the center of the beam due to blocks 1, . . . , j placed

completely within the intervals [0, x] and[x +

∑ni=j+1 li, L

]. Initially

D∗j (x) =

0 for x = 0, j = 0;

∞ otherwise

and the recursion for j = 1 to n is given by

D∗j (x) = min

D∗j−1(x− lj) + D(L/2, wj , lj , x− lj/2)

D∗j−1(x) + D(L/2, wj , lj , x +

∑ni=j lj − lj/2).

The optimal solution is then the sequence of blocks that minimizes D∗n(x) for

0 ≤ x ≤ L.

A similar dynamic programming recursion determines a sequence of blocks

that exactly minimizes the bending moment at the center of the beam, with

D∗j (x) replaced by M∗

j (x), the minimum bending moment at the center due to

blocks 1, . . . , j; and with D(L/2, wj , lj , rj) replaced by M(L/2, wj , lj , rj). The

optimal solution is the sequence of blocks that minimizes M∗n(x) for 0 ≤ x ≤ L.

Theorem 3. The dynamic programming recursions determine sequences of the

n blocks that minimize deflection or bending moment at the center of a beam of

length L. Furthermore, the recursions can be evaluated within O(nL) steps.

13

Proof. Each block j corresponds to a stage in the dynamic programming re-

cursion. The state of the process is given by the variable x (0 ≤ x ≤ L). For

any value of x, we consider two possible decisions: either place block Bj on

the left side (in the interval [x − lj , x]) or on the right side (in the interval

[x + (∑n

i=j+1 li), x + (∑n

i=j li)]). This determines the O(nL) time complexity,

and it dominates the O(n log n) time required to sort the blocks initially. The

correctness of the recursions follow from a straightforward application of the

Principle of Optimality [2].

By Lemma 2 any sequence that minimizes one of the criteria at the center

of the beam must be V-shaped and therefore cannot be “too bad” with respect

to the other criteria. Therefore, by Theorems 1 and 2 we have the following.

Corollary 2. Dynamic programming to minimize deflection at the center gives

a sequence that produces maximum deflection no greater than 16/9√

3 ≈ 1.027

times the theoretical minimum and bending moment no greater than 4 times

minimum.

Corollary 3. Dynamic programming to minimize bending moment at the center

gives a sequence that produces maximum deflection no greater than 32/9√

3 ≈

2.054 times the theoretical minimum and bending moment no greater than 2

times minimum.

To evaluate the recursion technically requires pseudo-polynomial time be-

cause the number of computational steps is a polynomial function of the length

L of the beam (rather than a binary encoding of L). In practice, when the

block lengths are not integral, the problem is converted to a discrete state space

problem by choosing an appropriate scale. For example, one can measure all

lengths in units of size gcd(l1, l2, . . . , ln) and the actual time complexity of the

dynamic program is O(nL/ gcd(l1, l2, . . . , ln)).

In the special case in which all blocks have equal length, then any two blocks

of a sequence can be interchanged without changing the position of any other

block. This allows a fully polynomial-time algorithm,with worst-case running

14

time independent of L. A simple interchange argument establishes that the

deflection and the bending moment at the center of the beam are both mini-

mized by sorting the blocks into non-decreasing order of weight and repeatedly

placing the next heaviest block as far from the center of the beam as possible.

This means that the maximum deflection will be within 16/9√

3 ≈ 1.027 times

the theoretical minimum and the maximum deflection will be within 2 times

minimum. This heuristic requires only O(n log n) time (for sorting the blocks).

6 Complexity

The following result shows that it is unlikely that either deflection or bending

moment can be minimized at the center of the beam any more quickly (in the

worst case) than in pseudo-polynomial time.

Theorem 4. The problems of minimizing the deflection and of minimizing the

bending moment at the center of a beam with homogeneous discrete loads are

NP-hard.

Proof. We will show that the decision problem for deflection is NP-complete.

The Deflection Problem can be stated as follows: Given a set of homogeneous

blocks, a beam, and a threshold value D0, is there a sequence of the blocks that

causes a deflection of at most D0 at the center of the beam? The reduction

is from the Partition Problem, which is known to be NP-complete [3]. An

instance of the Partition Problem is given by a set of indices J = 1, 2, . . . , n

and a set of positive integers {lj}j∈J ; the question is whether there exists a

partition J1, J2 such that∑

j∈J1lj =

∑j∈J2

lj . Given such an instance, create

an instance of the Deflection Problem as follows. There are n blocks, with

block j of length lj and weight lj , and there is an additional block of length

ln+1 = 1 and weight 0. The beam is of length L =∑n+1

j=1 lj ; the values of E

and I are irrelevant, so we take EI = 1 for simplicity. The threshold value is

D0 = (L−1)(5L3−3L2−3L+1)/384. The value of D0 is equal to the deflection

at the center of the beam due to 2 homogenous blocks of length (L− 1)/2 each

15

4 4•L/2

←−−−−−− (L− 1)/2 −−−−−−→←− 1 −→←−−−−−− (L− 1)/2 −−−−−−→

w/l = 1 w/l = 1

Figure 6: Partition Problem recast as a deflection problem

and of density w/l = 1 placed on each end of the beam as in Figure 6. If there

exists a partition J1, J2 of the set of indices, then the answer to the Deflection

Problem is affirmative: it suffices to place all the blocks corresponding to J1

next to each other on one end of the beam, and those corresponding to J2 on

the other end. The resulting placement is equivalent to that of Figure 6, and the

deflection at the center is exactly D0. If the answer to the Deflection Problem is

affirmative, the deflection at the center of the beam will necessarily be exactly

D0 and the blocks would have to be placed as in Figure 6, with block n + 1

exactly at the center of the beam. To see this, we consider any other placement

of the blocks in which block n+1 has its center offset from the center of the beam

by a distance δ and the other blocks are fitted in the intervals [0, (L− 1)/2 + δ]

and [(L + 1)/2 + δ, L]. We can compute the total deflection as if due to three

homogenous blocks corresponding to the regions of equal density. By simple

algebra the deflection at the center is larger than D0.

A similar reduction from the Partition Problem establishes the formal dif-

ficulty of minimizing bending moment. In this case we ask whether there is a

sequence of blocks with bending moment no greater than (L− 1)2/4.

Notice that this leaves open the question of whether maximum deflection

or maximum bending moment can be exactly minimized in pseudo-polynomial

time or whether these problems are “strongly” NP-hard [3]. The first alternative

would seem more likely if there is always an optimal sequence that is V-shaped

about some point (possibly not the center); however, we do not know whether

this is true.

16

7 Related work

One-dimensional problems of sequencing blocks have the same flavor as problems

of machine scheduling, only with an objective that is determined by physical

law rather than economics. For example, the special problems of minimizing the

deflection and bending moment at the center of a beam are similar to that of

scheduling n jobs on a single machine to minimize weighted absolute deviation

from a restrictive or small common due date [4, 5]. In the latter problem,

earliness costs are assessed against all jobs completed before the common due

date, and tardiness costs are assessed against all those completed after it. The

problem is to minimize the sum of these costs. The problems are analogous,

with the 1-dimensional beam corresponding to the line of time, the center of

the beam corresponding to the common due date, and deflection or bending

moment corresponding to earliness/tardiness costs. The lengths and weights of

the blocks correspond to the processing times and the economic weights of the

jobs, respectively. Lemma 2 establishes what Hall, Kubiak, and Sethi (1991)

refer to as the “weakly V-shaped property” of an optimal schedule [4]; and

the dynamic programming algorithm is a modification of the one presented by

Hoogeveen and van de Velde [5]. The difference between the problems is that for

the “earliness/tardiness” problem, an optimal schedule need not start at time

0, while for the deflection and the bending moment problems, the blocks are

confined to the interval [0,∑n

i=1 li].

8 Conclusions

We have suggested three heuristics to reduce maximum deflection and maximum

bending moment in a beam. These heuristics do not exactly minimize either

deflection or bending moment; but each heuristic has a performance guarantee

that says that neither deflection nor bending moment can be “too much” larger

than the minimum possible. Furthermore, the stronger the guarantee for one

objective, the weaker the guarantee for the other, as summarized in Table 1.

17

Error bound: Error bound:Heuristic max deflection max bending moment EffortArbitrary sequence ∞ ∞ O(1)V-shaped sequence 2.054 4 O(n log n)DP-deflection 1.017 4 O(nL)DP-bending moment 2.054 2 O(nL)

Table 1: Comparison of performance guarantees and computational effort forthree heuristics. The error bound is the largest possible ratio of the maximumdeflection (bending moment) to the smallest maximum deflection (bending mo-ment) possible.

To put these guarantees in perspective, note that for arbitrary sequences

of blocks there is no finite upper bound on the ratio of maximum deflection

to the minimum possible nor on the ratio of maximum bending moment to

the minimum possible. To see this, compare the sequences (B1, B3, B2) and

(B1, B2, B3), where B1 and B2 are both of length (L − ε)/2 and weight ε, and

B3 is of length ε and weight 1. As ε approaches 0, the ratios of maximum

deflections and of maximum bending moments become arbitrarily large.

We have given detailed analysis for the case of a beam with simple supports;

however our arguments apply when the beam has fixed supports at both ends.

Using the appropriately modified equations of deflection and bending moment,

we can show that for a beam with fixed supports at the ends, the deflection

at any point is at most 32/27 ≈ 1.185 times the deflection at the center of the

beam and the maximum bending moment is at most 4 times that at the center;

furthermore, these bounds are tight. Our previous analysis can be continued to

show that any V-shaped sequence causes deflection at most 64/27 ≈ 2.37 times

the theoretical minimum and bending moment at most 8 times the minimum.

Similarly, the dynamic program to minimize exactly deflection at the center gives

a sequence that causes deflection no more than 32/27 ≈ 1.185 times minimum;

and the dynamic program to minimize exactly bending moment at the center

gives a sequence that causes deflection no more than 4 times minimum.

It is worth remarking that an easily-solved special case with fixed supports

is the loading of a cantilever beam: The maximum deflection always occurs

18

at the free end of the beam, and the maximum bending moment at its fixed

end. A proof similar to that of Lemma 2 allows us to establish that the max-

imum deflection and the maximum bending moment of a cantilever beam are

both minimized by sorting the blocks in non-decreasing order of average density

and then repeatedly placing the next densest block as far from the free end as

possible. This requires only O(n log n) time, again for sorting the blocks.

The performance guarantees for our heuristics are weaker for the problem

of bending moment than for the problem of deflection, which suggests that the

problem of bending moment is in some sense more difficult. Unfortunately the

problem of bending moment is also probably the more keenly felt as a practical

problem. It would be useful as well as interesting to design heuristics with

improved performance guarantees for bending moment.

We have only considered the case of homogeneous blocks, for which deflection

and bending moment are each minimized at the center of the beam by some

sequence that has a V-shaped profile in the weight per unit length of the blocks.

For non-homogeneous blocks, the V-shape property does not hold, and no special

structure of the optimal solution is apparent. It is possible to use the same

heuristics to sequence a set of imaginary homogeneous blocks of the same weights

and lengths as the real blocks, then sequence the actual blocks in the same way

and orient them such that each block has its center of gravity farther from the

center of the beam. The worst-case performance of this procedure is not known

to the authors.

We have not considered other interesting structures such as beams with

differing end supports (for example, one simple and one fixed) or beams whose

supports are not at their ends. Also of interest are the 2-dimensional versions of

the problems, where it is desired to find an arrangement of blocks that minimizes

the deflection or the bending moment of an elastic plate.

The problems of minimizing deflection and bending moment in a beam are

examples of a more general class of problems that asks how a load should be

distributed on a given structure. This is complementary to the traditional ques-

tion of mechanical design, which asks for the structure to bear a given load.

19

Elsewhere we have suggested the name “combinatorial mechanics” for this ap-

parently new class of problems [1].

Acknowledgements

The authors were supported in part by the National Science Foundation (DDM-

9101581). In addition, J. Bartholdi was supported in part by the Office of Naval

Research (N00014-89-J-1571).

References

[1] Amiouny, S.V., Bartholdi, J.J. III, Vande Vate, J.H. and J. Zhang.

1992. “Balanced loading”, Operations Research 40(2):238–246.

[2] Bradley, S.P., Hax, A.C., and Magnanti, T.L. 1977. Applied Mathe-

matical Programming, Addison-Wesley.

[3] Garey, M.R. and D.S. Johnson. 1979. Computers and Intractability: A

Guide to the Theory of NP-Completeness, W. H. Freeman and Co., San

Francisco.

[4] Hall, N.G., W. Kubiak, and S.P. Sethi. 1991. “Earliness-tardiness

scheduling problems II: Deviation of completion times about a restrictive

common due date”, Operations Research 39(5):847–856.

[5] Hoogeveen, J.A. and S.L. van de Velde. 1989. “Scheduling around a

small common due date”, European Journal of Operational Research55:237–

242.

[6] Hopkins, B.R. 1987. Design Analysis of Shafts and Beams 2nd edition,

Krieger Publishing.

[7] Ku, Y.C. 1986. Deflection of Beams for Spans and Cross Sections, McGraw-

Hill.

20

[8] Shigley, J.S. and L.D. Mitchell. 1983. Mechanical Engineering Design

4th edition, McGraw-Hill.

21